Define the capability first#
Seeing task states, sending a reply, creating a new agent, selecting a model and coordinating dependencies are different actions. Each needs supported interfaces and clear authorization, rather than a single broad “controls all AI” claim.
Model choice also affects capability, latency, cost and data handling. Future routing should make those dependencies visible instead of hiding them behind an automatic switch.
Current useful scope#
Use CueSpec for documented task views and in-context replies in the Codex environment. Use the host’s own controls for task creation and model decisions. Several visible tasks can already be useful without being a fully orchestrated agent team.
A future integration guide should name actual providers, available operations, task identity rules and recovery when a destination is unavailable. Until that evidence exists, this page remains a direction with no delivery date or unlimited compatibility promise.